[NTLUG:Discuss] Re: LVM Span and data loss -- RAID-0, 1, 5, FRAID v. Intelligent RAID, etc...

Bryan J. Smith b.j.smith at ieee.org
Sun Nov 28 22:09:39 CST 2004


On Sun, 2004-11-28 at 22:16, Jay Urish wrote:
> Raid 0 is mirroring...

Just FYI (me being anal), I believe RAID-1 is mirroring.

RAID-0 is typically considered striping (although some refer to it as
spanning too) and non-redundant.

> What you need to do is find a raid controller with built in
> BIOPS like the one of the 3ware units..

Again, just FYI (me being anal), _every_ RAID card (and even non-
RAID ATA, SCSI, etc... cards) has 16-bit Int10h BIOS services so
it can boot.  The "FRAID" (free RAID) cards add in a bit of setup
logic.

Where the "FRAID" cards differ with either the 3Ware ASIC+SRAM
non-blocking I/O or traditional uC+DRAM buffering RAID cards is
the on-board intelligence.  I.e., once the OS loads, the FRAID
cards have 100% of their logic in the OS driver and use the main
CPU.  The ASIC+SRAM or uC+DRAM approaches have an on-board
intelligence with full embedded OS, SRAM/cache or DRAM/buffer to
handle queuing or cache/buffering operations, etc...

They also only require a minimal block driver which is typically
GPL because they do not expose any IP -- unlike the FRAID driver
which is typically licensed logic (although there is a generic,
"ataraid" GPL driver in the Linux kernel that can and is used with
the "hptraid", "pdcraid" and "silraid" per-FRAID card drivers, but
it is wholly unsupported by the vendors and may differ from their
on-disc organization).

For more on the 4 types of ATA RAID options:  
- Typical OS Logical Volume Management (LVM)
- Cheap Free RAID (FRAID) Cards
- Traditional uC+DRAM Intelligent RAID
- 3Ware non-blocking I/O ASIC+SRAM Intelligent RAID

See the article "Dissecting ATA RAID Options" in the 2004 April
issue of Sys Admin:  
  http://www.samag.com/articles/2004/0404/  

Unfortunately it is not a featured web article, so you need to
have a subscription or have bought it at the store.  Subscribers
get _all_ past issues on CD, just FYI.  ;->

> Then snag another 250gb drive and set up raid5

RAID-5 is _not_ necessarily better.  It is more efficient disk-wise,
and works like an N-1 (where N = number of disks) RAID-0 when it
comes to _reads_, but RAID-3, 4 and 5 are _much_slower_ when it comes
to writes than RAID-0, RAID-0+1 or even RAID-1 in some cases.

"Efficiency" is discussed in the "Dissecting ATA RAID Options" article
including a _detailed_ table on read v. write v. disk cost.  I guess
I could repost that table if someone wanted it.

FYI, the 3Ware Escalade 7000 and 8000 series with their small 1-4MB of
SRAM are _not_ ideal for RAID-5.  The small 0 wait state SRAM is perfect
for RAID-0, 1 and 0+1 operations, but for lots of random RAID-5 writes,
it can overflow and stall write operations.

Something like either the 3Ware Escalade 9000 series with 128MB of
SDRAM, or some of the new Intel IO331/332 (XScale) superscalar
microcontroller based ATA RAID controllers with 128MB of SDRAM would be
far more ideal for RAID-5.

In reality, I've found 4+ disc RAID-0+1 on a 3Ware Escalade 8506 series
to be the most ideal in price/performance.

Sometimes a combination is good as well.  E.g., using 3Ware Escalade
7506 and 8506 series products with 8 or 12 channels, I'll use 2-disc
RAID-1 or 4-disc RAID-0+1 for the "system" (especially swap, /tmp and
/var), and then 6-8 disc RAID-5 for the "data" partitions (especially
read-only databases or home directories).


-- 
Bryan J. Smith                                    b.j.smith at ieee.org 
-------------------------------------------------------------------- 
Subtotal Cost of Ownership (SCO) for Windows being less than Linux
Total Cost of Ownership (TCO) assumes experts for the former, costly
retraining for the latter, omitted "software assurance" costs in 
compatible desktop OS/apps for the former, no free/legacy reuse for
latter, and no basic security, patch or downtime comparison at all.





More information about the Discuss mailing list